socket: use accept4 when available

[panjf2000]

Linux, *BSD, and Solaris support accept4 system call that enables the caller to assign additional flags and save some extra system calls.
It can come in handy when O_NONBLOCK or/and FD_CLOEXEC is/are required on a socket after being accepted.

Ref:
https://man7.org/linux/man-pages/man2/accept.2.html https://man.freebsd.org/cgi/man.cgi?query=accept4
https://man.dragonflybsd.org/?command=accept&section=2 https://man.openbsd.org/accept.2
https://man.netbsd.org/accept.2
https://docs.oracle.com/cd/E88353_01/html/E37843/accept4-3c.html
https://www.gnu.org/software/gnulib/manual/html_node/accept4.html

[testclutch]

Analysis of PR #16979 at 745d7e4f:

Test 101 failed, which has NOT been flaky recently, so there could be a real issue in this PR.

Test 103 failed, which has NOT been flaky recently, so there could be a real issue in this PR.

Test 108 failed, which has NOT been flaky recently, so there could be a real issue in this PR.

Test 119 failed, which has NOT been flaky recently, so there could be a real issue in this PR.

Test 144 failed, which has NOT been flaky recently, so there could be a real issue in this PR.

Test 145 failed, which has NOT been flaky recently, so there could be a real issue in this PR.

Test 212 failed, which has NOT been flaky recently, so there could be a real issue in this PR.

There are more failures, but that's enough from Gha.

Generated by Testclutch

[panjf2000]

This is weird, I passed all the tests on my Linux server:

Anything special on those failed CIs? @bagder @icing @vszakats

[dfandrich]

The failing tests all seem to have debug mode enabled. Can you reproduce the failures if you configure with --enable-debug?

[panjf2000]

The failing tests all seem to have debug mode enabled. Can you reproduce the failures if you configure with --enable-debug?

[panjf2000]

Some failed tests arose, but they don't seem to be the same as those previous ones on CI's.

[bagder]

Add accept4 support to memdebug:

From a212acb9f2cfe740529aa03cd4fb173ec176367e Mon Sep 17 00:00:00 2001
From: Daniel Stenberg <daniel@haxx.se>
Date: Sun, 6 Apr 2025 11:21:42 +0200
Subject: [PATCH] memdebug: add accept4

---
 lib/memdebug.c | 18 ++++++++++++++++++
 lib/memdebug.h |  6 ++++++
 2 files changed, 24 insertions(+)

diff --git a/lib/memdebug.c b/lib/memdebug.c
index 58e4614f5f..b351726b32 100644
--- a/lib/memdebug.c
+++ b/lib/memdebug.c
@@ -376,10 +376,28 @@ curl_socket_t curl_dbg_accept(curl_socket_t s, void *saddr, void *saddrlen,
                  source, line, sockfd);
 
   return sockfd;
 }
 
+#ifdef HAVE_ACCEPT4
+curl_socket_t curl_dbg_accept4(curl_socket_t s, void *saddr, void *saddrlen,
+                               int flags,
+                               int line, const char *source)
+{
+  struct sockaddr *addr = (struct sockaddr *)saddr;
+  curl_socklen_t *addrlen = (curl_socklen_t *)saddrlen;
+
+  curl_socket_t sockfd = accept4(s, addr, addrlen, flags);
+
+  if(source && (sockfd != CURL_SOCKET_BAD))
+    curl_dbg_log("FD %s:%d accept() = %" FMT_SOCKET_T "\n",
+                 source, line, sockfd);
+
+  return sockfd;
+}
+#endif
+
 /* separate function to allow libcurl to mark a "faked" close */
 void curl_dbg_mark_sclose(curl_socket_t sockfd, int line, const char *source)
 {
   if(source)
     curl_dbg_log("FD %s:%d sclose(%" FMT_SOCKET_T ")\n",
diff --git a/lib/memdebug.h b/lib/memdebug.h
index aabcaadef4..fbea0c38bd 100644
--- a/lib/memdebug.h
+++ b/lib/memdebug.h
@@ -90,10 +90,13 @@ CURL_EXTERN void curl_dbg_mark_sclose(curl_socket_t sockfd,
                                       int line, const char *source);
 CURL_EXTERN int curl_dbg_sclose(curl_socket_t sockfd,
                                 int line, const char *source);
 CURL_EXTERN curl_socket_t curl_dbg_accept(curl_socket_t s, void *a, void *alen,
                                           int line, const char *source);
+CURL_EXTERN curl_socket_t curl_dbg_accept4(curl_socket_t s, void *saddr,
+                                           void *saddrlen, int flags,
+                                           int line, const char *source);
 #ifdef HAVE_SOCKETPAIR
 CURL_EXTERN int curl_dbg_socketpair(int domain, int type, int protocol,
                                     curl_socket_t socket_vector[2],
                                     int line, const char *source);
 #endif
@@ -157,10 +160,13 @@ CURL_EXTERN ALLOC_FUNC
 #define socket(domain,type,protocol)\
  curl_dbg_socket((int)domain, type, protocol, __LINE__, __FILE__)
 #undef accept /* for those with accept as a macro */
 #define accept(sock,addr,len)\
  curl_dbg_accept(sock, addr, len, __LINE__, __FILE__)
+#undef accept4 /* for those with accept as a macro */
+#define accept4(sock,addr,len,flags)                            \
+  curl_dbg_accept4(sock, addr, len, flags, __LINE__, __FILE__)
 #ifdef HAVE_SOCKETPAIR
 #define socketpair(domain,type,protocol,socket_vector)\
  curl_dbg_socketpair((int)domain, type, protocol, socket_vector, \
                      __LINE__, __FILE__)
 #endif
-- 
2.49.0

[panjf2000]

Any chance that this failed CI is a flaky one and is irrelevant to this PR?

[bagder]

Any chance that this failed CI is a flaky one and is irrelevant to this PR?

I am pretty sure that it is and is not result of your work.

[panjf2000]

Alright, I think we are good to go! Could you take a glance at this? @bagder

[bagder]

Thanks!

[Myriachan]

On Windows, you can get the atomic equivalent of O_CLOEXEC by calling WSASocketW() with WSA_FLAG_NO_HANDLE_INHERIT | WSA_FLAG_OVERLAPPED. (Note that socket() uses WSA_FLAG_OVERLAPPED internally, so this flag is needed if using WSASocketW() instead.)

WSA_FLAG_NO_HANDLE_INHERIT requires Windows 8.0+ though, so if you need 7 compatibility it'll have to detect versions.

WinSock's accept() copies the no-inherit flag from the listening socket, so there's no direct equivalent of accept4() needed.